Dementia: mortality prediction model

 Researchers developed a validated mortality prediction model for community-dwelling older adults with dementia (see dementia mortality predictor JamaIntMed2022 in dropbox, or doi:10.1001/jamainternmed.2022.4326)

 

Details: 

-- 4267 participants with probable dementia who were part of the Health and Retirement Study (HRS), a study of community-dwelling adults at least 65 years old, from 1998-2016 (this was the derivation cohort)

    -- patients were interviewed every two years to assess changes and outcomes, with proxy interviews as needed with a spouse or other family member

    -- probable dementia was defined by a previously validated algorithm, shown to have high accuracy when compared with the gold standard dementia diagnoses

    -- mean age 82 (42% at least age 85), 69% female, 75% white/12% Black/10% Hispanic, BMI <18.5 in 7%, 18.5-25 in 47%, 25-30 in 29%, > 30 in 13%; 52% never smokers/41% former/7% current, 39% married or partnered, 38% live alone, at least one ADL dependency in 33%/at least one IADL difficulty in 60%; difficulty walking several blocks 67%, vigorous physical activity 15%

        -- comorbidities: cancer 19%, diabetes 24%, heart disease 40%, hypertension 65%, lung disease 13%, stroke 24%

    -- median follow-up time was 3.9 years (interquartile range 2.0-6.8 years); 81% died at the end of follow-up

 

-- the National Health and Aging Trends Study (NHATS), a nationally representative longitudinal study of Medicare beneficiaries who were classified as having probable dementia based on defined cutoffs, from 2011- 2019 (this served as the external validation cohort: ie, the derived prediction model from the HRS study was applied to this cohort, comparing the predicted with the individuals' actual outcomes)

    -- mean age not reported, but 43% at least age 85; 59% female, 45% white/33% Black/14% Hispanic, BMI <18.5 in 5%, 18.5-25 in 39%, 25-30 in 29%, > 30 in 19%; 57% never smokers/36% former/6% current, 37% married or partnered, 25% live alone, at least one ADL dependency in 36%/at least one IADL difficulty in 78%; difficulty walking several blocks 67%, vigorous physical activity 15%

        -- comorbidities: cancer 15%, diabetes 33%, heart disease 32%, hypertension 74%, lung disease 21%, stroke 18%

 

--  included candidate predictors for mortality (ie, they assessed the HRS cohort for all of these, with the resultant prediction model honed down to the ones statistically associated with mortality):

    -- demographics: age category (five-year categories beginning at age 65 until >90yo), sex, and marital status

    -- health/behavioral factors: BMI (divided into <18.5, 18.5 to <25, 25 to <30, >30), smoking status (never smoker, former smoke, current smoker), participation in vigorous physical activity (hardly ever/never versus any vigorous physical activity)

    -- functional status: number of dependencies of ADLs (bathing, eating, dressing, getting in and out of bed, and toileting), difficulties with instrumental IADLs (preparing meals, managing medications, managing money, using a telephone, grocery shopping)

    -- comorbidities: cancer, diabetes, heart disease, lung disease, hypertension, stroke (these were basically the same from both the derivation and validation cohorts)

 

-- model evaluation: discrimination, measured as the integrated area under the receiver operating characteristic (ROC) curves over time; and calibration, defined as time-specific values at clinically relevant time points of 1, 2, 5, and 10 years, reflecting the level of agreement between the observed and predicted probability of mortality

 

-- Primary outcome: time until all-cause death

 

Results: 

-- Final prediction model for the HRS cohort, through the statistical analysis included:

    -- age, sex, BMI, smoking status, ADL dependency count, IADLs difficulty count, difficulty walking several blocks, participation in vigorous physical exercise, and chronic conditions

       -- the adjusted hazard ratio, the items most profoundly associated with death were age throughout the range, but exceeding a 35% increase by age 70 and increasing to a more than four-fold incidence in those over 90yo; BMI <18.5 had a 56% increase, current smokers a 37% increase, and lung disease conferred a 35% increased risk

 

-- AUC (integrated area under the curve) overall 0.76 (0.75-0.76) for HRS

    -- time specific AUC:

 

        -- at 1 year: 0.73 (0.70-0.75)

        -- at 5 years: 0.75 (0.73-0.77)

        -- at 10 years: 0.84 (0.82-0.85)

            -- ie the calibration assessment showed good agreement across a range of predicted risk from 1 to 10 years for HRS

-- for NHATS, the time-specific AUC values were:

        -- at 1 year: 0.73 (0.70-0.76)

        -- at 2 years: 0.72 (0.70-0.75)

        -- at 5 years: 0.74 (0.71-0.76)

            -- ie, remarkably similar to the above numbers for HRS (though no 10-year data for NHATS, but it did continue the increasing trend over time)

 

-- For illustration, here is the predicted timing to death for 10 randomly selected individuals with dementia from HRC within each decile of predicted risk:

 

 

 as above, a person 70-74 with lots of the measured problems has the worst prognosis, even lower than someone >90 who is pretty healthy

 

Commentary:                         

-- an estimated 6.5 million Americans >65yo are living with Alzheimer’s and related dementias in 2022

-- median survival after diagnosis of dementia varies from 3.3 to 11.7 years

-- having reasonable mortality predictions is important for an array of reasons: psychologically preparing the patient and family for the anticipated life expectancy, financial planning, advance care planning including potential needs for long-term care, clinical decision-making (when to stop preventive health maintenance, or medical interventions overall), etc

-- one risk calculator that is helpful is at eprognosis.com ( https://eprognosis.ucsf.edu/alexlee-result.php?language=English ), which asks 16 questions about the patient, finding, for example:

    -- a 70yo woman with BMI 30 and several medical problems and impaired ADLs has a 5-year predicted mortality of 52% and 10-yr of 89%

    -- a 90 yo woman with BMI 25 and more independent and healthier has a 5-year predicted mortality of 60% and 10-yr 94%

        -- ie, the mortality expectancy at 5 and 10 years of these 2 people is pretty similar despite their 20-year age difference.

        -- and, this is quite useful in helping patients and families, and guiding us, in determining what routine preventive services are appropriate, etc as above

        -- but, this eprognosis predictor does not even ask about dementia, so not relevant to the patients we see with dementia, as assessed in the above study

-- there have been a few other predictive models developed, but in general they are not for community-dwelling persons, do not have more than a few years of followup, and may not apply to US populations (some of the studies were done in Sweden or the Netherlands, where the issues may be somewhat different, since these countries have much better health care systems, which also focus more on providing a multitude of supports and services for the elderly)

 

-- The above study was of a large sample of community-dwelling elderly with dementia having up to 10-year predictions, and found a remarkable concordance of the derived model from the NRC cohort and the validation NHATS cohort, the first such study done

-- the above study was done in people with dementia, but it is important to realize that there are important non-pharmacologic approaches to decrease the risk of developing dementia in the first place, and there are  several small studies finding that cognitive rehabilitation and exercise programs may reduce the progression of dementia and decrease the likelihood of adverse events (eg falls/broken hips/etc) which could hasten accelerated  declines

    -- for prior blogs on lifestyle and dementia:

        -- see http://gmodestmedblogs.blogspot.com/2019/07/dementia-genetics-and-lifestyle-both.html which found that genetics and lifestyle both play a role in dementia

        -- and http://gmodestmedblogs.blogspot.com/2020/01/exercise-and-cognitive-health.html finding that moderate-to-vigorous exercise enhance the volume of the hippocampus, the area of the brain involved in consolidating information from short-term into long-term memory and one of the 1^st regions of the brain affected by dementia

 

Limitations:

-- the determination of dementia was done from existing databases without having lots of granular data, and some of the data collected was different in each of the databases (ie the 2 databases were not set up to run during the same time period and collecting the same data the same way)

    -- much of the information was binary and not as  continuous variables: eg having impairment of at least one ADL or IADL probably obscures the difference between those having just one and those with 5;  or vigorous exercise being none/hardly ever vs doing any does not really quantify well the potential benefits of exercise by quantity, and vigorous or not

-- no stratification by types of dementia, or many of the important psychosocial issues that might affect mental function: eg depression, stress, poor sleep, etc

-- no quantification of the degree of dementia by formal testing, and there are likely large differences in outcomes related to the dementia severity and its trajectory (stable for many months, slowly vs rapidly getting worse)

-- NHATS did not have 10-year outcomes, decreasing the validation of the predictive model to 5 years

 

So, seems like a useful tool to help us, patients, and families get important prognostic information to help guide the patient in their medical but especially in their life decisions and preparedness for the future. Longer-term followup and involving more granular data (as noted in the Limitations above) would likely improve the accuracy of the prediction model, though as it is it seemed to do quite well...

geoff

 

If you would like to be on the regular email list for upcoming blogs, please contact me at gmodest@uphams.org

to get access to all of the blogs (2 options):

1. go to http://gmodestmedblogs.blogspot.com/ to see them in reverse chronological order

2. click on 3 parallel lines top left, if you want to see blogs by category, then click on "labels" and choose a category​

3. or you can just click on the magnifying glass on top right, then  type in a name in the search box and get all the blogs with that name in them

 

or: go to https://www.bucommunitymedicine.org/ , a website from the Community Medicine section at Boston Medical Center.  This site does have a very searchable and accessible list of my blogs (though there have been a few that did not upload over the last year or two). but overall it is much easier to view blogs and displays more at a time.

 

please feel free to circulate this to others. also, if you send me their emails, i can add them to the list